Utility of Large-Scale Recipe Data in Food Computing

This article aims to look at the recipe data analysis from a critical perspective, offering the authors' own learning experience from successes and failures of the research process. The present recipe research has been limited by the availability of data, which in the case of recipes mostly consists of texts depicting a variety of ingredients. This has contributed to a better understanding of flavour formation and nutritional value of food but has not led further to establishing a corpus of healthy and unhealthy foods. Time-related cooking aspects have remained largely out of the present research's scope due to the difficulties in obtaining immediately analyzable data. The same goes for the recipe-relate research on food texture, color and other aspects. In this research the methodology of topic modelling has been applied to analyze recipes in North American and Mexican cuisines in order to highlight the core culinary themes within these two cuisines. Potential for result analysis, as well as its limitations, are also discussed. Topic models of agglomerated data can be helpful in further multisensory research, as they provide some insights into the colour, the flavour and, potentially, the texture of certain groups of dishes. It can be combined further on with social media sentiment analysis and other research methods to better grasp the human relationship with food.